gpu-unlock Windows port doubles FP16 on GeForce: RTX 4090 hits 331.6 TFLOPS

The open-source gpu-unlock tool removes a firmware cap that limits Tensor Core FP16/BF16 throughput on consumer GeForce cards to roughly half of professional GPUs built on the same silicon. The Windows port v0.3.1, released October 24, applies SS0/SS1 register writes via a UEFI app before the driver loads: RTX 4090 FP16 rose from 165.6 to 331.6 TFLOPS, RTX 3090 from 70.55 to 125.91 TFLOPS. The project drew about 6,800 GitHub stars in eight days; Nvidia has not commented.

• RTX 4090 FP16 rose from 165.6 to 331.6 TFLOPS
• RTX 3090 FP16 up from 70.55 to 125.91 TFLOPS (+78.5%)
• RTX 3090 Ti BF16 up from 87.5 to 157.4 TFLOPS (+79.9%)
• Windows v0.3.1 needs manual EFI selection at each boot